variable transformed with I() seen as string, and variable not found

[F041]

I have a model like that

lmc= lm(yc~I(x1)^2+log(x2), data=data)
So not built with a recipe like specified here

When I run the API, locally, I see

[
  {
    "I(x1)": "**string**",
    "log(x2)": 0
  }
]

but of course I have it as numeric.

For the second variable, I get

{
  "error": "500 - Internal server error",
  "message": "Error in eval(predvars, data, env): object 'x2' not found\n"
}

I don't feel sure what I posted belongs to an issue, since I stated I deviated from the recipe good practice. Maybe it can just lead to a specification in the documentation.

[juliasilge]

I think this would work if you used the tidymodels add_formula() interface:

library(tidymodels)
#> Warning: package 'broom' was built under R version 4.4.1
library(vetiver)
#> 
#> Attaching package: 'vetiver'
#> The following object is masked from 'package:tune':
#> 
#>     load_pkgs

wf <- 
    workflow() |> 
    add_model(linear_reg()) |> 
    add_formula(mpg ~ I(cyl) + log(disp)) |> 
    fit(data = mtcars)

vetiver_ptype(wf)
#> # A tibble: 0 × 2
#> # ℹ 2 variables: cyl <dbl>, disp <dbl>

^{Created on 2024-11-08 with reprex v2.1.1}

You can definitely use lm() directly with a simple formula, like this example, but if you start using more complex formula syntax then lm() doesn't give you a way to get out the true inputs. For example:

library(vetiver)
mod <- lm(mpg ~ I(cyl) + log(disp), data = mtcars)
vetiver_ptype(mod)
#> # A tibble: 0 × 2
#> # ℹ 2 variables: I(cyl) <I<dbl>>, log(disp) <dbl>

^{Created on 2024-11-08 with reprex v2.1.1}

I will look for a way to clarify in the documentation these limitations of the formula interface.